add test-after-edit eval condition

[orban]

Summary

Adds a fifth experimental condition test_after_edit to the eval harness. The condition injects a preamble that forces the agent to run relevant tests after every source-file edit before making further changes, isolating the effect of tight test-feedback loops from context-injection effects (none/flat_llm/intent_layer).

Wired through:

• Condition enum (task_runner.py)
• preamble map in TaskRunner._build_prompt
• YAML-tasks condition list in run() (cli.py)
• reporter display label (reporter.py)
• new preamble string (prompt_builder.py)

Test plan

• pytest eval-harness/tests/test_task_runner.py::test_condition_enum
• Full run against one repo to confirm reports render the new condition label correctly

Notes

• Dropped an unrelated bash -lc → sh -c revert in docker_runner.py that had slipped into the same working tree — that change would have reintroduced exit 127 for repos using uv (see commit 9ab8632).
• Replaces add test-after-edit eval condition #35, which picked up 17 unrelated commits from nightshift/bus-factor.